blob: d538dd53710460b2f3a8b388ac62d5b5f47511f8 [file] [log] [blame]
mbligh6203ace2007-10-04 21:54:24 +00001#!/usr/bin/python -u
mbligh1ffd5dc2008-11-25 13:24:05 +00002# Copyright 2007-2008 Martin J. Bligh <mbligh@google.com>, Google Inc.
mbligh82648e52008-11-20 16:54:25 +00003# Released under the GPL v2
mblighdcd57a82007-07-11 23:06:47 +00004
mblighc8949b82007-07-23 16:33:58 +00005"""
Aviv Keshetde6bb192013-01-30 16:17:22 -08006Run a control file through the server side engine
mblighdcd57a82007-07-11 23:06:47 +00007"""
mbligh1ffd5dc2008-11-25 13:24:05 +00008
Fang Deng042c1472014-10-23 13:56:41 -07009import datetime
Paul Hobbs20cc72a2016-08-30 16:57:05 -070010import contextlib
Fang Deng042c1472014-10-23 13:56:41 -070011import getpass
12import logging
13import os
14import re
15import signal
Dan Shicf4d2032015-03-12 15:04:21 -070016import socket
Fang Deng042c1472014-10-23 13:56:41 -070017import sys
18import traceback
19import time
20import urllib2
mbligh1ffd5dc2008-11-25 13:24:05 +000021
Prathmesh Prabhua5eecda2016-11-23 16:48:40 -080022
mblighf5427bb2008-04-09 15:55:57 +000023import common
Dan Shi4f8c0242017-07-07 15:34:49 -070024from autotest_lib.client.bin.result_tools import utils as result_utils
25from autotest_lib.client.bin.result_tools import view as result_view
Dan Shia1ecd5c2013-06-06 11:21:31 -070026from autotest_lib.client.common_lib import control_data
Dan Shi32649b82015-08-29 20:53:36 -070027from autotest_lib.client.common_lib import error
Dan Shia1ecd5c2013-06-06 11:21:31 -070028from autotest_lib.client.common_lib import global_config
Allen Lif146e872017-08-15 18:24:31 -070029from autotest_lib.server import results_mocker
Prathmesh Prabhua5eecda2016-11-23 16:48:40 -080030
Dan Shi5e2efb72017-02-07 11:40:23 -080031try:
32 from chromite.lib import metrics
33except ImportError:
34 metrics = utils.metrics_mock
Prathmesh Prabhua5eecda2016-11-23 16:48:40 -080035
Dan Shia06f3e22015-09-03 16:15:15 -070036_CONFIG = global_config.global_config
37
Dan Shia1ecd5c2013-06-06 11:21:31 -070038
Jakob Jueliche497b552014-09-23 19:11:59 -070039# Number of seconds to wait before returning if testing mode is enabled
Prashanth B6285f6a2014-05-08 18:01:27 -070040TESTING_MODE_SLEEP_SECS = 1
Jakob Jueliche497b552014-09-23 19:11:59 -070041
mbligh9ff89cd2009-09-03 20:28:17 +000042
Kevin Cheng9b6930f2016-07-20 14:57:15 -070043from autotest_lib.server import frontend
showard75cdfee2009-06-10 17:40:41 +000044from autotest_lib.server import server_logging_config
showard043c62a2009-06-10 19:48:57 +000045from autotest_lib.server import server_job, utils, autoserv_parser, autotest
Dan Shia1ecd5c2013-06-06 11:21:31 -070046from autotest_lib.server import utils as server_utils
Paul Hobbs20cc72a2016-08-30 16:57:05 -070047from autotest_lib.server import site_utils
Kevin Chengadc99f92016-07-20 08:21:58 -070048from autotest_lib.server.cros.dynamic_suite import frontend_wrappers
Dan Shicf4d2032015-03-12 15:04:21 -070049from autotest_lib.site_utils import job_directories
Fang Deng042c1472014-10-23 13:56:41 -070050from autotest_lib.site_utils import job_overhead
Dan Shicf4d2032015-03-12 15:04:21 -070051from autotest_lib.site_utils import lxc
Ben Kwa966db082017-06-05 14:17:23 -070052from autotest_lib.site_utils.lxc import utils as lxc_utils
showard75cdfee2009-06-10 17:40:41 +000053from autotest_lib.client.common_lib import pidfile, logging_manager
mbligh92c0fc22008-11-20 16:52:23 +000054
Paul Hobbs20cc72a2016-08-30 16:57:05 -070055
Dan Shicf4d2032015-03-12 15:04:21 -070056# Control segment to stage server-side package.
57STAGE_SERVER_SIDE_PACKAGE_CONTROL_FILE = server_job._control_segment_path(
58 'stage_server_side_package')
59
Dan Shia06f3e22015-09-03 16:15:15 -070060# Command line to start servod in a moblab.
61START_SERVOD_CMD = 'sudo start servod BOARD=%s PORT=%s'
62STOP_SERVOD_CMD = 'sudo stop servod'
63
Alex Millerf1af17e2013-01-09 22:50:32 -080064def log_alarm(signum, frame):
65 logging.error("Received SIGALARM. Ignoring and continuing on.")
Alex Miller0528d6f2013-01-11 10:49:48 -080066 sys.exit(1)
Alex Millerf1af17e2013-01-09 22:50:32 -080067
Dan Shicf4d2032015-03-12 15:04:21 -070068
69def _get_machines(parser):
70 """Get a list of machine names from command line arg -m or a file.
71
72 @param parser: Parser for the command line arguments.
73
74 @return: A list of machine names from command line arg -m or the
75 machines file specified in the command line arg -M.
76 """
77 if parser.options.machines:
78 machines = parser.options.machines.replace(',', ' ').strip().split()
79 else:
80 machines = []
81 machines_file = parser.options.machines_file
82 if machines_file:
83 machines = []
84 for m in open(machines_file, 'r').readlines():
85 # remove comments, spaces
86 m = re.sub('#.*', '', m).strip()
87 if m:
88 machines.append(m)
89 logging.debug('Read list of machines from file: %s', machines_file)
90 logging.debug('Machines: %s', ','.join(machines))
91
92 if machines:
93 for machine in machines:
94 if not machine or re.search('\s', machine):
95 parser.parser.error("Invalid machine: %s" % str(machine))
96 machines = list(set(machines))
97 machines.sort()
98 return machines
99
100
Prathmesh Prabhu588007d2017-06-15 00:31:31 -0700101def _stage_ssp(parser, resultsdir):
Dan Shicf4d2032015-03-12 15:04:21 -0700102 """Stage server-side package.
103
104 This function calls a control segment to stage server-side package based on
105 the job and autoserv command line option. The detail implementation could
106 be different for each host type. Currently, only CrosHost has
107 stage_server_side_package function defined.
108 The script returns None if no server-side package is available. However,
109 it may raise exception if it failed for reasons other than artifact (the
110 server-side package) not found.
111
112 @param parser: Command line arguments parser passed in the autoserv process.
Prathmesh Prabhu588007d2017-06-15 00:31:31 -0700113 @param resultsdir: Folder to store results. This could be different from
114 parser.options.results: parser.options.results can be set to None
115 for results to be stored in a temp folder. resultsdir can be None
116 for autoserv run requires no logging.
Dan Shicf4d2032015-03-12 15:04:21 -0700117
Dan Shi14de7622016-08-22 11:09:06 -0700118 @return: (ssp_url, error_msg), where
119 ssp_url is a url to the autotest server-side package. None if
120 server-side package is not supported.
121 error_msg is a string indicating the failures. None if server-
122 side package is staged successfully.
Dan Shicf4d2032015-03-12 15:04:21 -0700123 """
Kevin Chengadc99f92016-07-20 08:21:58 -0700124 machines_list = _get_machines(parser)
Prathmesh Prabhu588007d2017-06-15 00:31:31 -0700125 machines_list = server_job.get_machine_dicts(machines_list, resultsdir,
126 parser.options.lab,
127 parser.options.host_attributes)
Kevin Chengadc99f92016-07-20 08:21:58 -0700128
Dan Shi36cfd832014-10-10 13:38:51 -0700129 # If test_source_build is not specified, default to use server-side test
130 # code from build specified in --image.
Kevin Chengadc99f92016-07-20 08:21:58 -0700131 namespace = {'machines': machines_list,
Dan Shi36cfd832014-10-10 13:38:51 -0700132 'image': (parser.options.test_source_build or
133 parser.options.image),}
Dan Shicf4d2032015-03-12 15:04:21 -0700134 script_locals = {}
135 execfile(STAGE_SERVER_SIDE_PACKAGE_CONTROL_FILE, namespace, script_locals)
Dan Shi14de7622016-08-22 11:09:06 -0700136 return script_locals['ssp_url'], script_locals['error_msg']
Dan Shicf4d2032015-03-12 15:04:21 -0700137
138
Dan Shiafa63872016-02-23 15:32:31 -0800139def _run_with_ssp(job, container_name, job_id, results, parser, ssp_url,
Dan Shi3be35af2016-08-25 23:22:40 -0700140 job_folder, machines):
Dan Shicf4d2032015-03-12 15:04:21 -0700141 """Run the server job with server-side packaging.
142
Dan Shi37befda2015-12-07 13:16:56 -0800143 @param job: The server job object.
Dan Shicf4d2032015-03-12 15:04:21 -0700144 @param container_name: Name of the container to run the test.
145 @param job_id: ID of the test job.
146 @param results: Folder to store results. This could be different from
147 parser.options.results:
148 parser.options.results can be set to None for results to be
149 stored in a temp folder.
150 results can be None for autoserv run requires no logging.
151 @param parser: Command line parser that contains the options.
152 @param ssp_url: url of the staged server-side package.
Dan Shiafa63872016-02-23 15:32:31 -0800153 @param job_folder: Name of the job result folder.
Dan Shi3be35af2016-08-25 23:22:40 -0700154 @param machines: A list of machines to run the test.
Dan Shicf4d2032015-03-12 15:04:21 -0700155 """
156 bucket = lxc.ContainerBucket()
157 control = (parser.args[0] if len(parser.args) > 0 and parser.args[0] != ''
158 else None)
Dan Shi37befda2015-12-07 13:16:56 -0800159 try:
Dan Shi3be35af2016-08-25 23:22:40 -0700160 dut_name = machines[0] if len(machines) >= 1 else None
Dan Shi37befda2015-12-07 13:16:56 -0800161 test_container = bucket.setup_test(container_name, job_id, ssp_url,
Dan Shiafa63872016-02-23 15:32:31 -0800162 results, control=control,
Dan Shi3be35af2016-08-25 23:22:40 -0700163 job_folder=job_folder,
164 dut_name=dut_name)
Dan Shi37befda2015-12-07 13:16:56 -0800165 except Exception as e:
166 job.record('FAIL', None, None,
167 'Failed to setup container for test: %s. Check logs in '
168 'ssp_logs folder for more details.' % e)
169 raise
170
Dan Shicf4d2032015-03-12 15:04:21 -0700171 args = sys.argv[:]
172 args.remove('--require-ssp')
Dan Shi77b79a62015-07-29 16:22:05 -0700173 # --parent_job_id is only useful in autoserv running in host, not in
174 # container. Include this argument will cause test to fail for builds before
175 # CL 286265 was merged.
176 if '--parent_job_id' in args:
177 index = args.index('--parent_job_id')
178 args.remove('--parent_job_id')
179 # Remove the actual parent job id in command line arg.
180 del args[index]
Dan Shicf4d2032015-03-12 15:04:21 -0700181
182 # A dictionary of paths to replace in the command line. Key is the path to
183 # be replaced with the one in value.
184 paths_to_replace = {}
185 # Replace the control file path with the one in container.
186 if control:
187 container_control_filename = os.path.join(
188 lxc.CONTROL_TEMP_PATH, os.path.basename(control))
189 paths_to_replace[control] = container_control_filename
190 # Update result directory with the one in container.
Dan Shi65374e22016-09-15 16:14:05 -0700191 container_result_dir = os.path.join(lxc.RESULT_DIR_FMT % job_folder)
Dan Shicf4d2032015-03-12 15:04:21 -0700192 if parser.options.results:
Dan Shicf4d2032015-03-12 15:04:21 -0700193 paths_to_replace[parser.options.results] = container_result_dir
194 # Update parse_job directory with the one in container. The assumption is
195 # that the result folder to be parsed is always the same as the results_dir.
196 if parser.options.parse_job:
Dan Shicf4d2032015-03-12 15:04:21 -0700197 paths_to_replace[parser.options.parse_job] = container_result_dir
198
199 args = [paths_to_replace.get(arg, arg) for arg in args]
200
201 # Apply --use-existing-results, results directory is aready created and
202 # mounted in container. Apply this arg to avoid exception being raised.
203 if not '--use-existing-results' in args:
204 args.append('--use-existing-results')
205
206 # Make sure autoserv running in container using a different pid file.
207 if not '--pidfile-label' in args:
208 args.extend(['--pidfile-label', 'container_autoserv'])
209
Dan Shid1f51232015-04-18 00:29:14 -0700210 cmd_line = ' '.join(["'%s'" % arg if ' ' in arg else arg for arg in args])
Dan Shicf4d2032015-03-12 15:04:21 -0700211 logging.info('Run command in container: %s', cmd_line)
Dan Shi37bee222015-04-13 15:46:47 -0700212 success = False
Dan Shicf4d2032015-03-12 15:04:21 -0700213 try:
214 test_container.attach_run(cmd_line)
Dan Shi37bee222015-04-13 15:46:47 -0700215 success = True
Dan Shi9d3454e2015-12-08 09:16:08 -0800216 except Exception as e:
217 # If the test run inside container fails without generating any log,
218 # write a message to status.log to help troubleshooting.
219 debug_files = os.listdir(os.path.join(results, 'debug'))
220 if not debug_files:
221 job.record('FAIL', None, None,
222 'Failed to run test inside the container: %s. Check '
223 'logs in ssp_logs folder for more details.' % e)
224 raise
Dan Shicf4d2032015-03-12 15:04:21 -0700225 finally:
Prathmesh Prabhua5eecda2016-11-23 16:48:40 -0800226 metrics.Counter(
227 'chromeos/autotest/experimental/execute_job_in_ssp').increment(
228 fields={'success': success})
Dan Shicf4d2032015-03-12 15:04:21 -0700229 test_container.destroy()
230
231
Dan Shi3f1b8a52015-04-21 11:11:06 -0700232def correct_results_folder_permission(results):
233 """Make sure the results folder has the right permission settings.
234
235 For tests running with server-side packaging, the results folder has the
236 owner of root. This must be changed to the user running the autoserv
237 process, so parsing job can access the results folder.
238 TODO(dshi): crbug.com/459344 Remove this function when test container can be
239 unprivileged container.
240
241 @param results: Path to the results folder.
242
243 """
244 if not results:
245 return
246
Aviv Keshetc03de792017-07-18 14:24:31 -0700247 utils.run('sudo -n chown -R %s "%s"' % (os.getuid(), results))
248 utils.run('sudo -n chgrp -R %s "%s"' % (os.getgid(), results))
Dan Shi3f1b8a52015-04-21 11:11:06 -0700249
250
Dan Shia06f3e22015-09-03 16:15:15 -0700251def _start_servod(machine):
252 """Try to start servod in moblab if it's not already running or running with
253 different board or port.
254
255 @param machine: Name of the dut used for test.
256 """
257 if not utils.is_moblab():
258 return
259
Dan Shi1cded882015-09-23 16:52:26 -0700260 logging.debug('Trying to start servod.')
Dan Shia06f3e22015-09-03 16:15:15 -0700261 try:
Kevin Cheng9b6930f2016-07-20 14:57:15 -0700262 afe = frontend.AFE()
Dan Shia06f3e22015-09-03 16:15:15 -0700263 board = server_utils.get_board_from_afe(machine, afe)
264 hosts = afe.get_hosts(hostname=machine)
265 servo_host = hosts[0].attributes.get('servo_host', None)
266 servo_port = hosts[0].attributes.get('servo_port', 9999)
267 if not servo_host in ['localhost', '127.0.0.1']:
Dan Shi1cded882015-09-23 16:52:26 -0700268 logging.warn('Starting servod is aborted. The dut\'s servo_host '
269 'attribute is not set to localhost.')
Dan Shia06f3e22015-09-03 16:15:15 -0700270 return
271 except (urllib2.HTTPError, urllib2.URLError):
272 # Ignore error if RPC failed to get board
273 logging.error('Failed to get board name from AFE. Start servod is '
274 'aborted')
275 return
276
277 try:
278 pid = utils.run('pgrep servod').stdout
279 cmd_line = utils.run('ps -fp %s' % pid).stdout
280 if ('--board %s' % board in cmd_line and
281 '--port %s' % servo_port in cmd_line):
282 logging.debug('Servod is already running with given board and port.'
283 ' There is no need to restart servod.')
284 return
285 logging.debug('Servod is running with different board or port. '
286 'Stopping existing servod.')
287 utils.run('sudo stop servod')
288 except error.CmdError:
289 # servod is not running.
290 pass
291
292 try:
293 utils.run(START_SERVOD_CMD % (board, servo_port))
294 logging.debug('Servod is started')
295 except error.CmdError as e:
296 logging.error('Servod failed to be started, error: %s', e)
297
298
Dan Shic68fefb2015-04-07 10:10:52 -0700299def run_autoserv(pid_file_manager, results, parser, ssp_url, use_ssp):
Dan Shicf4d2032015-03-12 15:04:21 -0700300 """Run server job with given options.
301
302 @param pid_file_manager: PidFileManager used to monitor the autoserv process
303 @param results: Folder to store results.
304 @param parser: Parser for the command line arguments.
305 @param ssp_url: Url to server-side package.
Dan Shic68fefb2015-04-07 10:10:52 -0700306 @param use_ssp: Set to True to run with server-side packaging.
Dan Shicf4d2032015-03-12 15:04:21 -0700307 """
Dan Shiec1d47d2015-02-13 11:38:13 -0800308 if parser.options.warn_no_ssp:
Dan Shic68fefb2015-04-07 10:10:52 -0700309 # Post a warning in the log.
Dan Shiec1d47d2015-02-13 11:38:13 -0800310 logging.warn('Autoserv is required to run with server-side packaging. '
311 'However, no drone is found to support server-side '
312 'packaging. The test will be executed in a drone without '
313 'server-side packaging supported.')
314
jadmanski0afbb632008-06-06 21:10:57 +0000315 # send stdin to /dev/null
316 dev_null = os.open(os.devnull, os.O_RDONLY)
317 os.dup2(dev_null, sys.stdin.fileno())
318 os.close(dev_null)
mblighdbf37612007-11-24 19:38:11 +0000319
Dan Shie8aeb662016-06-30 11:22:03 -0700320 # Create separate process group if the process is not a process group
321 # leader. This allows autoserv process to keep running after the caller
322 # process (drone manager call) exits.
323 if os.getpid() != os.getpgid(0):
324 os.setsid()
mbligh1d42d4e2007-11-05 22:42:00 +0000325
Dan Shicf4d2032015-03-12 15:04:21 -0700326 # Container name is predefined so the container can be destroyed in
327 # handle_sigterm.
328 job_or_task_id = job_directories.get_job_id_or_task_id(
329 parser.options.results)
330 container_name = (lxc.TEST_CONTAINER_NAME_FMT %
Dan Shid68d51c2015-04-21 17:00:42 -0700331 (job_or_task_id, time.time(), os.getpid()))
Dan Shiafa63872016-02-23 15:32:31 -0800332 job_folder = job_directories.get_job_folder_name(parser.options.results)
Dan Shicf4d2032015-03-12 15:04:21 -0700333
jadmanski0afbb632008-06-06 21:10:57 +0000334 # Implement SIGTERM handler
mblighc2299562009-07-02 19:00:36 +0000335 def handle_sigterm(signum, frame):
Simran Basi9d9b7292013-10-16 16:44:07 -0700336 logging.debug('Received SIGTERM')
mblighff7d61f2008-12-22 14:53:35 +0000337 if pid_file_manager:
338 pid_file_manager.close_file(1, signal.SIGTERM)
Simran Basi49e21e62013-10-17 12:40:33 -0700339 logging.debug('Finished writing to pid_file. Killing process.')
Dan Shi3f1b8a52015-04-21 11:11:06 -0700340
341 # Update results folder's file permission. This needs to be done ASAP
342 # before the parsing process tries to access the log.
343 if use_ssp and results:
344 correct_results_folder_permission(results)
345
Simran Basid6b83772014-01-06 16:31:30 -0800346 # TODO (sbasi) - remove the time.sleep when crbug.com/302815 is solved.
347 # This sleep allows the pending output to be logged before the kill
348 # signal is sent.
349 time.sleep(.1)
Dan Shic68fefb2015-04-07 10:10:52 -0700350 if use_ssp:
Dan Shicf4d2032015-03-12 15:04:21 -0700351 logging.debug('Destroy container %s before aborting the autoserv '
352 'process.', container_name)
353 try:
354 bucket = lxc.ContainerBucket()
355 container = bucket.get(container_name)
356 if container:
357 container.destroy()
358 else:
359 logging.debug('Container %s is not found.', container_name)
360 except:
361 # Handle any exception so the autoserv process can be aborted.
Dan Shi65374e22016-09-15 16:14:05 -0700362 logging.exception('Failed to destroy container %s.',
363 container_name)
Dan Shie4a4f9f2015-07-20 09:00:25 -0700364 # Try to correct the result file permission again after the
365 # container is destroyed, as the container might have created some
366 # new files in the result folder.
367 if results:
368 correct_results_folder_permission(results)
Dan Shicf4d2032015-03-12 15:04:21 -0700369
jadmanski0afbb632008-06-06 21:10:57 +0000370 os.killpg(os.getpgrp(), signal.SIGKILL)
mblighfaf0cd42007-11-19 16:00:24 +0000371
jadmanski0afbb632008-06-06 21:10:57 +0000372 # Set signal handler
mblighc2299562009-07-02 19:00:36 +0000373 signal.signal(signal.SIGTERM, handle_sigterm)
mbligha46678d2008-05-01 20:00:01 +0000374
Simran Basid6b83772014-01-06 16:31:30 -0800375 # faulthandler is only needed to debug in the Lab and is not avaliable to
376 # be imported in the chroot as part of VMTest, so Try-Except it.
377 try:
378 import faulthandler
379 faulthandler.register(signal.SIGTERM, all_threads=True, chain=True)
380 logging.debug('faulthandler registered on SIGTERM.')
381 except ImportError:
Christopher Grant4beca022015-06-16 15:14:47 -0400382 sys.exc_clear()
Simran Basid6b83772014-01-06 16:31:30 -0800383
David Rochberg8a60d1e2011-02-01 14:22:07 -0500384 # Ignore SIGTTOU's generated by output from forked children.
385 signal.signal(signal.SIGTTOU, signal.SIG_IGN)
386
Alex Millerf1af17e2013-01-09 22:50:32 -0800387 # If we received a SIGALARM, let's be loud about it.
388 signal.signal(signal.SIGALRM, log_alarm)
389
mbligha5f5e542009-12-30 16:57:49 +0000390 # Server side tests that call shell scripts often depend on $USER being set
391 # but depending on how you launch your autotest scheduler it may not be set.
392 os.environ['USER'] = getpass.getuser()
393
mblighb2bea302008-07-24 20:25:57 +0000394 label = parser.options.label
mbligh374f3412009-05-13 21:29:45 +0000395 group_name = parser.options.group_name
mblighb2bea302008-07-24 20:25:57 +0000396 user = parser.options.user
397 client = parser.options.client
398 server = parser.options.server
jadmanski0afbb632008-06-06 21:10:57 +0000399 install_before = parser.options.install_before
mblighb2bea302008-07-24 20:25:57 +0000400 install_after = parser.options.install_after
401 verify = parser.options.verify
402 repair = parser.options.repair
showard45ae8192008-11-05 19:32:53 +0000403 cleanup = parser.options.cleanup
Alex Millercb79ba72013-05-29 14:43:00 -0700404 provision = parser.options.provision
Dan Shi07e09af2013-04-12 09:31:29 -0700405 reset = parser.options.reset
Alex Miller667b5f22014-02-28 15:33:39 -0800406 job_labels = parser.options.job_labels
mblighb2bea302008-07-24 20:25:57 +0000407 no_tee = parser.options.no_tee
jadmanski0afbb632008-06-06 21:10:57 +0000408 parse_job = parser.options.parse_job
mblighe7d9c602009-07-02 19:02:33 +0000409 execution_tag = parser.options.execution_tag
410 if not execution_tag:
411 execution_tag = parse_job
jadmanski0afbb632008-06-06 21:10:57 +0000412 ssh_user = parser.options.ssh_user
413 ssh_port = parser.options.ssh_port
414 ssh_pass = parser.options.ssh_pass
jadmanskidef0c3c2009-03-25 20:07:10 +0000415 collect_crashinfo = parser.options.collect_crashinfo
mblighe0cbc912010-03-11 18:03:07 +0000416 control_filename = parser.options.control_filename
Scott Zawalski91493c82013-01-25 16:15:20 -0500417 test_retry = parser.options.test_retry
beepscb6f1e22013-06-28 19:14:10 -0700418 verify_job_repo_url = parser.options.verify_job_repo_url
Christopher Wileyf594c5e2013-07-03 18:25:30 -0700419 skip_crash_collection = parser.options.skip_crash_collection
Aviv Keshet18ee3142013-08-12 15:01:51 -0700420 ssh_verbosity = int(parser.options.ssh_verbosity)
Fang Deng6cc20de2013-09-06 15:47:32 -0700421 ssh_options = parser.options.ssh_options
Dan Shib669cbd2013-09-13 11:17:17 -0700422 no_use_packaging = parser.options.no_use_packaging
Simran Basi1bf60eb2015-12-01 16:39:29 -0800423 host_attributes = parser.options.host_attributes
424 in_lab = bool(parser.options.lab)
Paul Hobbs68d98592017-08-22 02:22:49 -0700425 cloud_trace_context = parser.options.cloud_trace_context
mbligha46678d2008-05-01 20:00:01 +0000426
mblighb2bea302008-07-24 20:25:57 +0000427 # can't be both a client and a server side test
428 if client and server:
Eric Li861b2d52011-02-04 14:50:35 -0800429 parser.parser.error("Can not specify a test as both server and client!")
mblighb2bea302008-07-24 20:25:57 +0000430
Alex Millercb79ba72013-05-29 14:43:00 -0700431 if provision and client:
432 parser.parser.error("Cannot specify provisioning and client!")
433
434 is_special_task = (verify or repair or cleanup or collect_crashinfo or
Dan Shi07e09af2013-04-12 09:31:29 -0700435 provision or reset)
Alex Millercb79ba72013-05-29 14:43:00 -0700436 if len(parser.args) < 1 and not is_special_task:
Eric Li861b2d52011-02-04 14:50:35 -0800437 parser.parser.error("Missing argument: control file")
mbligha46678d2008-05-01 20:00:01 +0000438
Aviv Keshet18ee3142013-08-12 15:01:51 -0700439 if ssh_verbosity > 0:
440 # ssh_verbosity is an integer between 0 and 3, inclusive
441 ssh_verbosity_flag = '-' + 'v' * ssh_verbosity
Fang Dengd1c2b732013-08-20 12:59:46 -0700442 else:
443 ssh_verbosity_flag = ''
Aviv Keshet18ee3142013-08-12 15:01:51 -0700444
showard45ae8192008-11-05 19:32:53 +0000445 # We have a control file unless it's just a verify/repair/cleanup job
jadmanski0afbb632008-06-06 21:10:57 +0000446 if len(parser.args) > 0:
447 control = parser.args[0]
448 else:
449 control = None
mbligha46678d2008-05-01 20:00:01 +0000450
Dan Shicf4d2032015-03-12 15:04:21 -0700451 machines = _get_machines(parser)
mbligh374f3412009-05-13 21:29:45 +0000452 if group_name and len(machines) < 2:
Dan Shicf4d2032015-03-12 15:04:21 -0700453 parser.parser.error('-G %r may only be supplied with more than one '
454 'machine.' % group_name)
mbligh374f3412009-05-13 21:29:45 +0000455
Christopher Wiley8a91f232013-07-09 11:02:27 -0700456 kwargs = {'group_name': group_name, 'tag': execution_tag,
Dan Shicf4d2032015-03-12 15:04:21 -0700457 'disable_sysinfo': parser.options.disable_sysinfo}
Dan Shi70647ca2015-07-16 22:52:35 -0700458 if parser.options.parent_job_id:
459 kwargs['parent_job_id'] = int(parser.options.parent_job_id)
mblighe0cbc912010-03-11 18:03:07 +0000460 if control_filename:
461 kwargs['control_filename'] = control_filename
Simran Basi1bf60eb2015-12-01 16:39:29 -0800462 if host_attributes:
463 kwargs['host_attributes'] = host_attributes
464 kwargs['in_lab'] = in_lab
jadmanski0afbb632008-06-06 21:10:57 +0000465 job = server_job.server_job(control, parser.args[1:], results, label,
466 user, machines, client, parse_job,
Fang Dengd1c2b732013-08-20 12:59:46 -0700467 ssh_user, ssh_port, ssh_pass,
Aviv Keshetc5947fa2013-09-04 14:06:29 -0700468 ssh_verbosity_flag, ssh_options,
469 test_retry, **kwargs)
Dan Shicf4d2032015-03-12 15:04:21 -0700470
showard75cdfee2009-06-10 17:40:41 +0000471 job.logging.start_logging()
mbligh4608b002010-01-05 18:22:35 +0000472 job.init_parser()
mbligha46678d2008-05-01 20:00:01 +0000473
mbligh161fe6f2008-06-19 16:26:04 +0000474 # perform checks
475 job.precheck()
476
jadmanski0afbb632008-06-06 21:10:57 +0000477 # run the job
478 exit_code = 0
Dan Shic1b8bdd2015-09-14 23:11:24 -0700479 auto_start_servod = _CONFIG.get_config_value(
480 'AUTOSERV', 'auto_start_servod', type=bool, default=False)
Paul Hobbs20cc72a2016-08-30 16:57:05 -0700481
Prathmesh Prabhu27bba962017-01-24 15:13:07 -0800482 site_utils.SetupTsMonGlobalState('autoserv', indirect=False,
Paul Hobbs66251f62017-08-22 02:31:07 -0700483 auto_flush=False, short_lived=True)
jadmanski0afbb632008-06-06 21:10:57 +0000484 try:
Prathmesh Prabhu27bba962017-01-24 15:13:07 -0800485 try:
486 if repair:
487 if auto_start_servod and len(machines) == 1:
488 _start_servod(machines[0])
489 job.repair(job_labels)
490 elif verify:
491 job.verify(job_labels)
492 elif provision:
493 job.provision(job_labels)
494 elif reset:
495 job.reset(job_labels)
496 elif cleanup:
497 job.cleanup(job_labels)
498 else:
499 if auto_start_servod and len(machines) == 1:
500 _start_servod(machines[0])
501 if use_ssp:
502 try:
503 _run_with_ssp(job, container_name, job_or_task_id,
504 results, parser, ssp_url, job_folder,
505 machines)
506 finally:
507 # Update the ownership of files in result folder.
508 correct_results_folder_permission(results)
Dan Shicf4d2032015-03-12 15:04:21 -0700509 else:
Prathmesh Prabhu27bba962017-01-24 15:13:07 -0800510 if collect_crashinfo:
511 # Update the ownership of files in result folder. If the
512 # job to collect crashinfo was running inside container
513 # (SSP) and crashed before correcting folder permission,
514 # the result folder might have wrong permission setting.
Dan Shiafa63872016-02-23 15:32:31 -0800515 try:
516 correct_results_folder_permission(results)
Prathmesh Prabhu27bba962017-01-24 15:13:07 -0800517 except:
518 # Ignore any error as the user may not have root
519 # permission to run sudo command.
520 pass
Aviv Keshet92bf7b62017-02-13 15:34:03 -0800521 metric_name = ('chromeos/autotest/experimental/'
522 'autoserv_job_run_duration')
523 f = {'in_container': utils.is_in_container(),
524 'success': False}
525 with metrics.SecondsTimer(metric_name, fields=f) as c:
526 job.run(install_before, install_after,
527 verify_job_repo_url=verify_job_repo_url,
528 only_collect_crashinfo=collect_crashinfo,
529 skip_crash_collection=skip_crash_collection,
530 job_labels=job_labels,
531 use_packaging=(not no_use_packaging))
532 c['success'] = True
533
Prathmesh Prabhu27bba962017-01-24 15:13:07 -0800534 finally:
Hidehiko Abe06893302017-06-24 07:32:38 +0900535 job.close()
Dan Shiffd5b822017-07-14 11:16:23 -0700536 # Special task doesn't run parse, so result summary needs to be
537 # built here.
538 if results and (repair or verify or reset or cleanup or provision):
Dan Shi4f8c0242017-07-07 15:34:49 -0700539 # Throttle the result on the server side.
540 try:
541 result_utils.execute(
542 results, control_data.DEFAULT_MAX_RESULT_SIZE_KB)
543 except:
544 logging.exception(
545 'Non-critical failure: Failed to throttle results '
546 'in directory %s.', results)
547 # Build result view and report metrics for result sizes.
Dan Shiffd5b822017-07-14 11:16:23 -0700548 site_utils.collect_result_sizes(results)
jadmanski0afbb632008-06-06 21:10:57 +0000549 except:
jadmanski27b37ea2008-10-29 23:54:31 +0000550 exit_code = 1
jadmanski0afbb632008-06-06 21:10:57 +0000551 traceback.print_exc()
Prathmesh Prabhu27bba962017-01-24 15:13:07 -0800552 finally:
553 metrics.Flush()
mbligha46678d2008-05-01 20:00:01 +0000554
mblighff7d61f2008-12-22 14:53:35 +0000555 if pid_file_manager:
556 pid_file_manager.num_tests_failed = job.num_tests_failed
557 pid_file_manager.close_file(exit_code)
jadmanskie0dffc32008-12-15 17:30:30 +0000558 job.cleanup_parser()
showard21baa452008-10-21 00:08:39 +0000559
jadmanski27b37ea2008-10-29 23:54:31 +0000560 sys.exit(exit_code)
mbligha46678d2008-05-01 20:00:01 +0000561
562
Paul Hobbs68d98592017-08-22 02:22:49 -0700563def record_autoserv(options, start_time):
Fang Deng042c1472014-10-23 13:56:41 -0700564 """Record autoserv end-to-end time in metadata db.
565
566 @param options: parser options.
Paul Hobbs68d98592017-08-22 02:22:49 -0700567 @param start_time: When autoserv started
Fang Deng042c1472014-10-23 13:56:41 -0700568 """
569 # Get machine hostname
570 machines = options.machines.replace(
571 ',', ' ').strip().split() if options.machines else []
572 num_machines = len(machines)
573 if num_machines > 1:
574 # Skip the case where atomic group is used.
575 return
576 elif num_machines == 0:
577 machines.append('hostless')
578
579 # Determine the status that will be reported.
Paul Hobbs68d98592017-08-22 02:22:49 -0700580 status = get_job_status(options)
581 is_special_task = status not in [
582 job_overhead.Status.RUNNING, job_overhead.Status.GATHERING]
583 job_or_task_id = job_directories.get_job_id_or_task_id(options.results)
584 duration_secs = (datetime.datetime.now() - start_time).total_seconds()
585 job_overhead.record_state_duration(
586 job_or_task_id, machines[0], status, duration_secs,
587 is_special_task=is_special_task)
588
589
590def get_job_status(options):
591 """Returns the HQE Status for this run.
592
593 @param options: parser options.
594 """
Fang Deng042c1472014-10-23 13:56:41 -0700595 s = job_overhead.STATUS
596 task_mapping = {
597 'reset': s.RESETTING, 'verify': s.VERIFYING,
598 'provision': s.PROVISIONING, 'repair': s.REPAIRING,
599 'cleanup': s.CLEANING, 'collect_crashinfo': s.GATHERING}
Paul Hobbs68d98592017-08-22 02:22:49 -0700600 match = [task for task in task_mapping if getattr(options, task, False)]
601 return task_mapping[match[0]] if match else s.RUNNING
Fang Deng042c1472014-10-23 13:56:41 -0700602
603
mbligha46678d2008-05-01 20:00:01 +0000604def main():
Fang Deng042c1472014-10-23 13:56:41 -0700605 start_time = datetime.datetime.now()
jadmanski0afbb632008-06-06 21:10:57 +0000606 # grab the parser
607 parser = autoserv_parser.autoserv_parser
mbligha5cb4062009-02-17 15:53:39 +0000608 parser.parse_args()
mbligha46678d2008-05-01 20:00:01 +0000609
jadmanski0afbb632008-06-06 21:10:57 +0000610 if len(sys.argv) == 1:
611 parser.parser.print_help()
612 sys.exit(1)
mbligha6f13082008-06-05 23:53:46 +0000613
showard75cdfee2009-06-10 17:40:41 +0000614 if parser.options.no_logging:
615 results = None
616 else:
617 results = parser.options.results
mbligh80e1eba2008-11-19 00:26:18 +0000618 if not results:
619 results = 'results.' + time.strftime('%Y-%m-%d-%H.%M.%S')
Dan Shi14de7622016-08-22 11:09:06 -0700620 results = os.path.abspath(results)
showard566d3c02010-01-12 18:57:01 +0000621 resultdir_exists = False
622 for filename in ('control.srv', 'status.log', '.autoserv_execute'):
623 if os.path.exists(os.path.join(results, filename)):
624 resultdir_exists = True
mbligh4608b002010-01-05 18:22:35 +0000625 if not parser.options.use_existing_results and resultdir_exists:
mbligh80e1eba2008-11-19 00:26:18 +0000626 error = "Error: results directory already exists: %s\n" % results
627 sys.stderr.write(error)
628 sys.exit(1)
mbligha788dc42009-03-26 21:10:16 +0000629
630 # Now that we certified that there's no leftover results dir from
631 # previous jobs, lets create the result dir since the logging system
632 # needs to create the log file in there.
633 if not os.path.isdir(results):
634 os.makedirs(results)
showard75cdfee2009-06-10 17:40:41 +0000635
Prathmesh Prabhu7ae68ae2017-06-15 00:29:31 -0700636 # If the job requires to run with server-side package, try to stage server-
637 # side package first. If that fails with error that autotest server package
638 # does not exist, fall back to run the job without using server-side
639 # packaging. If option warn_no_ssp is specified, that means autoserv is
640 # running in a drone does not support SSP, thus no need to stage server-side
641 # package.
642 ssp_url = None
643 ssp_url_warning = False
644 if (not parser.options.warn_no_ssp and parser.options.require_ssp):
Prathmesh Prabhu588007d2017-06-15 00:31:31 -0700645 ssp_url, ssp_error_msg = _stage_ssp(parser, results)
Prathmesh Prabhu7ae68ae2017-06-15 00:29:31 -0700646 # The build does not have autotest server package. Fall back to not
647 # to use server-side package. Logging is postponed until logging being
648 # set up.
649 ssp_url_warning = not ssp_url
650
Dan Shic68fefb2015-04-07 10:10:52 -0700651 # Server-side packaging will only be used if it's required and the package
652 # is available. If warn_no_ssp is specified, it means that autoserv is
653 # running in a drone does not have SSP supported and a warning will be logs.
654 # Therefore, it should not run with SSP.
655 use_ssp = (not parser.options.warn_no_ssp and parser.options.require_ssp
656 and ssp_url)
657 if use_ssp:
Dan Shie28de552015-05-06 16:51:58 -0700658 log_dir = os.path.join(results, 'ssp_logs') if results else None
Dan Shicf4d2032015-03-12 15:04:21 -0700659 if log_dir and not os.path.exists(log_dir):
660 os.makedirs(log_dir)
661 else:
662 log_dir = results
Dan Shi3f1b8a52015-04-21 11:11:06 -0700663
showard75cdfee2009-06-10 17:40:41 +0000664 logging_manager.configure_logging(
Dan Shicf4d2032015-03-12 15:04:21 -0700665 server_logging_config.ServerLoggingConfig(),
666 results_dir=log_dir,
showard10d84172009-06-18 23:16:50 +0000667 use_console=not parser.options.no_tee,
668 verbose=parser.options.verbose,
669 no_console_prefix=parser.options.no_console_prefix)
Dan Shicf4d2032015-03-12 15:04:21 -0700670
Dan Shi0b754c52015-04-20 14:20:38 -0700671 if ssp_url_warning:
672 logging.warn(
673 'Autoserv is required to run with server-side packaging. '
674 'However, no server-side package can be found based on '
Dan Shi6450e142016-03-11 11:52:20 -0800675 '`--image`, host attribute job_repo_url or host OS version '
676 'label. It could be that the build to test is older than the '
677 'minimum version that supports server-side packaging. The test '
Dan Shi14de7622016-08-22 11:09:06 -0700678 'will be executed without using erver-side packaging. '
679 'Following is the detailed error:\n%s', ssp_error_msg)
Dan Shi0b754c52015-04-20 14:20:38 -0700680
showard75cdfee2009-06-10 17:40:41 +0000681 if results:
mbligha788dc42009-03-26 21:10:16 +0000682 logging.info("Results placed in %s" % results)
mbligh10717632008-11-19 00:21:57 +0000683
mbligh4608b002010-01-05 18:22:35 +0000684 # wait until now to perform this check, so it get properly logged
Dan Shicf4d2032015-03-12 15:04:21 -0700685 if (parser.options.use_existing_results and not resultdir_exists and
Dan Shiff78f112015-06-12 13:34:02 -0700686 not utils.is_in_container()):
mbligh4608b002010-01-05 18:22:35 +0000687 logging.error("No existing results directory found: %s", results)
688 sys.exit(1)
689
Dan Shicf4d2032015-03-12 15:04:21 -0700690 logging.debug('autoserv is running in drone %s.', socket.gethostname())
Aviv Keshet5c40ec62013-08-20 12:11:12 -0700691 logging.debug('autoserv command was: %s', ' '.join(sys.argv))
mbligh4608b002010-01-05 18:22:35 +0000692
Dan Shicf4d2032015-03-12 15:04:21 -0700693 if parser.options.write_pidfile and results:
mbligh4608b002010-01-05 18:22:35 +0000694 pid_file_manager = pidfile.PidFileManager(parser.options.pidfile_label,
695 results)
jadmanskid5ab8c52008-12-03 16:27:07 +0000696 pid_file_manager.open_file()
mblighff7d61f2008-12-22 14:53:35 +0000697 else:
698 pid_file_manager = None
mbligha46678d2008-05-01 20:00:01 +0000699
Allen Lid5abdab2017-02-07 16:03:43 -0800700 autotest.Autotest.set_install_in_tmpdir(
jadmanskif22fea82008-11-26 20:57:07 +0000701 parser.options.install_in_tmpdir)
702
jadmanski0afbb632008-06-06 21:10:57 +0000703 exit_code = 0
Allen Lif146e872017-08-15 18:24:31 -0700704 # TODO(beeps): Extend this to cover different failure modes.
705 # Testing exceptions are matched against labels sent to autoserv. Eg,
706 # to allow only the hostless job to run, specify
707 # testing_exceptions: test_suite in the shadow_config. To allow both
708 # the hostless job and dummy_Pass to run, specify
709 # testing_exceptions: test_suite,dummy_Pass. You can figure out
710 # what label autoserv is invoked with by looking through the logs of a test
711 # for the autoserv command's -l option.
712 testing_exceptions = _CONFIG.get_config_value(
713 'AUTOSERV', 'testing_exceptions', type=list, default=[])
714 test_mode = _CONFIG.get_config_value(
715 'AUTOSERV', 'testing_mode', type=bool, default=False)
716 test_mode = (results_mocker and test_mode and not
717 any([ex in parser.options.label
718 for ex in testing_exceptions]))
719 is_task = (parser.options.verify or parser.options.repair or
720 parser.options.provision or parser.options.reset or
721 parser.options.cleanup or parser.options.collect_crashinfo)
jadmanski0afbb632008-06-06 21:10:57 +0000722 try:
723 try:
Allen Lif146e872017-08-15 18:24:31 -0700724 if test_mode:
725 # The parser doesn't run on tasks anyway, so we can just return
726 # happy signals without faking results.
727 if not is_task:
728 machine = parser.options.results.split('/')[-1]
729
730 # TODO(beeps): The proper way to do this would be to
731 # refactor job creation so we can invoke job.record
732 # directly. To do that one needs to pipe the test_name
733 # through run_autoserv and bail just before invoking
734 # the server job. See the comment in
735 # puppylab/results_mocker for more context.
736 results_mocker.ResultsMocker(
737 'unknown-test', parser.options.results, machine
738 ).mock_results()
739 return
740 else:
741 run_autoserv(pid_file_manager, results, parser, ssp_url,
742 use_ssp)
Aviv Keshet5c40ec62013-08-20 12:11:12 -0700743 except SystemExit as e:
jadmanski0afbb632008-06-06 21:10:57 +0000744 exit_code = e.code
Aviv Keshet5c40ec62013-08-20 12:11:12 -0700745 if exit_code:
Aviv Keshet5ae0a002017-05-05 10:23:33 -0700746 logging.exception('Uncaught SystemExit with code %s', exit_code)
747 except Exception:
jadmanski0afbb632008-06-06 21:10:57 +0000748 # If we don't know what happened, we'll classify it as
749 # an 'abort' and return 1.
Aviv Keshet5ae0a002017-05-05 10:23:33 -0700750 logging.exception('Uncaught Exception, exit_code = 1.')
jadmanski0afbb632008-06-06 21:10:57 +0000751 exit_code = 1
752 finally:
mblighff7d61f2008-12-22 14:53:35 +0000753 if pid_file_manager:
754 pid_file_manager.close_file(exit_code)
Fang Deng042c1472014-10-23 13:56:41 -0700755 # Record the autoserv duration time. Must be called
756 # just before the system exits to ensure accuracy.
Paul Hobbs68d98592017-08-22 02:22:49 -0700757 record_autoserv(parser.options, start_time)
jadmanski0afbb632008-06-06 21:10:57 +0000758 sys.exit(exit_code)
mblighfaf0cd42007-11-19 16:00:24 +0000759
mblighbb421852008-03-11 22:36:16 +0000760
mbligha46678d2008-05-01 20:00:01 +0000761if __name__ == '__main__':
jadmanski0afbb632008-06-06 21:10:57 +0000762 main()